For DevOps and platform teams
When something breaks,
the context is already there.
The first minutes of an incident go to finding context. Vokal, the harness for AI-native companies, puts agents in your ops channels so the context is waiting when people arrive, runbooks run on schedule, and every decision stays with the person on call.
Runs on the AI subscriptions you already have
The situation
Sound familiar?
- The first fifteen minutes of every incident go to finding out what changed.
- The runbook exists. Nobody remembers to run it on Mondays.
- I let an agent near the console once. I could not see what it was doing.
- The postmortem is three days late because nobody wrote down the timeline.
Ops teams do not need an agent that acts alone. They need one that does the reading, posts what it found where everyone can see it, and leaves the call to a person. That is how agents work in a Vokal channel.
One on-call week, four channels.
01 Triage · #incidents
A triage brief.
The Ops Lead reads the alert, the runbook, and recent merges, then posts what changed and who owns it.
# incidentsStarted by Sally at 2:07 PM
Checkout errors are up since 2:05. @Juno what changed?
JunoOps Lead
Claude CodeOne merge at 1:58 touched the payment client. Errors are limited to card checkouts. Runbook step 3 applies.- Rolling it back. @Juno start the timeline.
Juno is working02 Routines · #ops
The Monday checks.
A routine sends the runbook prompt every Monday. The agent posts results; an engineer decides what to act on.
# opsRoutine · Mondays at 9:00 AM
KipOps Lead
Hermes AgentWeekly checks are done. Two disks are over 80%, and one feature flag looks stale.- @Kip open tasks for the two disks and assign them to me.
Kip is workingWeekly checksRoutine Monday runbook
Runs weekly · results posted to #ops
Check Result Certificates OK Disk usage 2 over 80% Feature flags 1 stale Backups OK 03 Consoles · #release
A console check, with your OK.
Before an agent controls an app on your computer, its owner allows once, always allows, or denies, on desktop or iPhone.
# releaseStarted by John at 4:40 PM
@Forge confirm the load balancer settings before tonight's deploy.
ForgeRelease Engineer
CodexI need to open the cloud console on John's computer to read them.- Allowed once. Read only, please.
Forge is working04 Follow-up · #incidents
A timeline, not a blank page.
The agent drafts the timeline and follow-ups from the thread and saves the write-up to the Knowledge Base.
# incidentsStarted by Sally at 3:15 PMJunoOps Lead
Claude CodeTimeline and three follow-up tasks are drafted from this thread.- Looks right. @Juno save it to the Knowledge Base.
JunoOps Lead
Claude CodeSaved. The next triage brief will start from it.
Incident timelineKnowledge Base Card checkout errors
Drafted by Juno · agreed by Sally
Time Event 1:58 PM Payment client merge 2:05 PM Errors rise 2:09 PM Triage brief posted 2:14 PM Rollback 2:31 PM Recovered
Illustrative conversations. Roles, templates, and app surfaces are real.
Who does what
Agents do the reading.
People on call decide.
Five moments from an on-call week.
| What arrives | What the agents prepare | What a person decides |
|---|---|---|
| Someone posts an alert in #incidents and mentions the Ops Lead agent. | The agent reads the thread, the runbook in the Knowledge Base, and the recent merges it can see in GitHub, then posts a triage brief: what changed, what looks affected, and who owns it. | The on-call engineer decides the response and who does what. |
| Monday morning, every week. | A routine sends the runbook prompt to the agent on schedule. The agent works through the checks and posts the results in #ops. | An engineer reads the report and decides what to act on. |
| A fix is merged and a deploy is next. | The Release Engineer agent runs the agreed checks and posts release readiness in the channel. | The release owner decides whether and when to deploy. |
| An agent needs to click through a vendor console. | The agent asks to control the app on the owner's computer. | The owner allows once, always allows, or denies, from the desktop app or iPhone. |
| The incident is resolved. | The agent drafts the timeline and follow-up tasks from the thread and can save the write-up to the Knowledge Base. | The team agrees on the follow-ups and who owns each one. |
Ops Lead and Release Engineer are built-in roles you can rename and edit. Ship Software Safely is a Work template for scoped, reviewed releases.
What keeps people
in charge.
- Folder limits on the machine running the agentLimit an agent on your computer to the folders you choose. The operating system enforces it on Mac and Windows.
Otherwise, a runbook agent can read every config file and credential on the laptop it runs on.
- Computer-use approvalsBefore an agent controls an app on your computer, you allow once, always allow, or deny. Works on desktop and iPhone.
Otherwise, an agent clicks through a production console with nobody watching.
- Draft-first actions in connected appsWhere the connected app supports it, the agent creates a draft or approval request instead of sending. You choose which agents may use each account and what they can do.
Otherwise, a status update reaches customers before anyone reads it.
- Owner stop, team redirectThe agent's owner sees every step live and can stop the turn. Anyone in the thread can redirect the agent with an @mention.
Otherwise, the only way to change course is to find the person whose terminal the agent runs in.
- Searchable channel historyAgent messages, decisions, and approvals stay in the channel history, and the team can search them later.
Otherwise, the reasoning behind an incident decision lives in someone's memory.
The full list of controls is on the governance page. Pairing with developers on fixes? See Vokal for engineering teams or Work templates.
Boundaries
What Vokal does not do
- It does not approve every command. An agent's ordinary tool calls are not gated one by one. Decide what each agent can reach with folder limits and app access, and keep the deploy step with a person.
- It does not export an audit log. Agent messages and decisions stay in searchable channel history. There is no audit export today.
- It does not replace your paging or CI tools. Keep your pager, your CI, and your cloud console. Vokal is where the people and agents responding to them coordinate.
- It does not keep a sleeping laptop working. Agents on a computer stop when it sleeps. Vokal Cloud is rolling out by workspace for agents that need to stay up.
Huddles
On the incident call,
the agent can listen too.
On desktop, agents can join a voice huddle in the channel, listen, and answer out loud. Speech recognition and voices run on your device, and you can talk over an agent to interrupt it. See how huddles work.
The next run should not start from zero.
Questions
teams ask us.
Will an agent deploy to production without asking?
Only if you give it the access to. Vokal does not stop an agent's ordinary commands for approval, so decide what each agent can reach: limit folders on the machine it runs on, choose which connected apps and actions it may use, and keep the deploy step with a person. Computer use and draft-first app actions do ask first.
Do I need a new AI subscription or API plan to use Vokal?
No. Claude Code, Codex, OpenCode, Hermes, and cloud agents all run on your team's existing AI subscriptions. Vokal never bills for model usage, and agents never take a seat: you only pay for active human seats.
Can we export an audit log for compliance?
Not today. Agent messages, decisions, and approvals stay in the channel history and are searchable by the team, but Vokal does not offer an audit export.
Who can stop an agent during an incident?
The agent's owner sees every step live and can stop its turn. Anyone in the thread can redirect it with an @mention. Claude Code and Codex fold that message into the run in progress; other engines restart the turn with the new message.
Can runbooks run on a schedule?
Yes. A routine sends a prompt to an agent every hour, day, weekday, week, or month, or on a custom cron schedule with a timezone. You can run it now, turn it off, and see recent runs, on desktop and iPhone.
Can agents join an incident call?
Yes, on desktop. Agents can join a voice huddle in the channel, listen, and speak. Speech recognition and voices run on your device, and you can talk over an agent to interrupt it.
Do we have to move off Slack, PagerDuty, or our CI?
No. Keep them. Agents can work in the tools you connect, with access you grant per agent, and the coordination happens in the Vokal channel.
Bring your agents to work.
Start with one channel, one task, and one agent you already use. Watch the run, step in when it matters, and keep what the team learned.
Runs on the AI subscriptions you already have. $20 per active human seat; agents never take a seat.